3/3 Use it in your agent code:
agent = ComputerAgent(model="cua/anthropic/claude-opus-4.5", tools=[computer] )
All usage tracked in your dashboard with live cost breakdowns.
Try it free → cua.ai/dashboard/playground
1
391
18 Nov 2025
oui.
agent = ComputerAgent(
model="anthropic/claude-sonnet-4-5-20250929",
tools=[computer],
trajectory_dir="trajectories",
use_prompt_caching=True,
max_trajectory_budget=10.0,
)
use trajectories
1
2
105
25 Oct 2025
JUST IN: Learn to build a computer-use agent using local AI models. Tutorial covers creating a virtual desktop, tool interface, reasoning engine, and interaction capabilities. Explore full codes for detailed implementation. #AI #ComputerAgent
2
52
agent = ComputerAgent("huggingface-local/ByteDance-Seed/UI-TARS-2")
███████▒▒▒ 70%
4 Sep 2025
We can finally share UI-TARS-2🥳🥳 — a native GUI agent trained with multi-turn agent RL
⚡️⚡️Key highlights (all-in-one model!):
💻Computer Use: 47.5 OSWorld · 50.6 WindowsAgentArena
📱Phone Use: 73.3 AndroidWorld
🛜Browser Use: 88.2% Online-Mind2Web
🎮Gameplay: ~60% human on 15 titles · strong on LMGame-Bench
🧑💻TerminalUse: 68.7 SWE-Bench · 45.3 TerminalBench
🔨Tool Use: 29.6 BrowseComp
Hybrid flows: GUI clicks terminal cmds API calls in one trace
Paper arxiv.org/abs/2509.02544
Demo seed-tars.com/showcase/ui-ta…
4
5
55
14,411
But here's the thing: you can combine models by specializing them. Put a grounding model (good at seeing/clicking) with a planning model (good at reasoning) using :
agent = ComputerAgent(
model="huggingface-local/HelloKKMe/GTA1-7B openai/gpt-4o",
tools=[computer]
)
3/5
1
3
106
The problem: GUI models are all different. Some want pixel coordinates, others want percentages, some use weird tokens like <|loc095|>. We solved this with a universal interface that works the same across Anthropic, OpenAI, Hugging Face, and more:
agent = ComputerAgent(
model="anthropic/claude-3-5-sonnet-20241022",
tools=[computer]
)
2/5
1
3
119
Once your computer is running, pointing an agent at it is just as simple:
from agent import ComputerAgent
agent = ComputerAgent("openrouter/z-ai/glm-4.5v", tools=[computer])
async for _ in agent. run("Click on the search bar and type 'hello world'"):
pass
This turns the container into a controlled, browser-accessible desktop that your model can drive.
4/5
1
2
309
How we run it in Cua (new composed agents - blog post soon):
- Reasoning model: OpenAI GPT-5
- Grounding model: Salesforce GTA1-7B
- Action space: CUA Cloud Instances (macOS/Linux/Windows)
agent = ComputerAgent(
model="huggingface-local/HelloKKMe/GTA1-7B openai/gpt-5",
tools=[computer]
)
docs.trycua.com/docs/agent-s…
2/2
1
16
896
agent = ComputerAgent(..)
await agent. run("Where's our pizza? /cc @supportingden")
1
5
223
29 May 2025
Now we have managed Cue cloud, like:
computer = Computer(
api_key=os.getenv("CUA_API_KEY"),
provider_type=VMProviderType .CLOUD,
)
agent = ComputerAgent(computer=computer,
loop=AgentLoop.OPENAI)
Launching public preview of C/ua Cloud Containers today - Docker for Computer-Use Agents. Zero local setup. Same Computer and Agent interfaces. Scale 1-100 agents instantly.
1/6
4
629
Getting started is ridiculously easy:
computer = Computer(
api_key=os.getenv("CUA_API_KEY"),
provider_type=VMProviderType.CLOUD,
)
agent = ComputerAgent(computer=computer,
loop=AgentLoop.OPENAI)
# That's it. You're in the cloud.
5/6
1
11
1,080
